Cross-time-domain prediction method for inclination trend of side wall of deep and large foundation pit

Abstract

The invention discloses a cross-time-domain prediction method for the inclination trend of a side wall of a deep and large foundation pit. The method comprises the following steps: S1, obtaining native displacement data and historical influence data based on influence factor data of the inclination degree of the side wall of the foundation pit and inclination displacement data of the side wall of the foundation pit; s2, obtaining a trained native inclination displacement prediction model; s3, obtaining a trained disturbance inclination displacement prediction model; and S4, realizing foundation pit side wall displacement prediction based on the total displacement prediction model. According to the method, the general situation of on-site foundation pit displacement is calculated by using the original inclination displacement prediction model, and then the comprehensive influence of subsequent external influence factors on the foundation pit displacement is predicted by using the disturbance inclination displacement prediction model; therefore, excessive protection and collapse accidents commonly existing in the foundation pit construction process can be scientifically and effectively avoided, manpower and material resources of site construction are saved, meanwhile, site construction safety can be guaranteed, and the practical application requirements of the construction site are better met.

Description

Technical Field

[0001] The invention relates to the technical field of foundation pit stability prediction, and in particular to a cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit. Background Art

[0002] At present, in the field of foundation pit construction, the analysis and prediction of the inclination deformation displacement of the foundation pit side wall has become one of the important bases for judging whether the construction results are stable or not. However, most of the existing analysis methods evaluate the stability of the foundation pit based on the on-site soil deformation detection results combined with manual experience. This method not only consumes a lot of manpower and material resources, but also has unsatisfactory evaluation accuracy, which may eventually lead to over-protection or collapse accidents.

[0003] In fact, the deformation of the sidewall of the foundation pit is caused by the combined effects of basic geological conditions and various external factors (such as excavation depth (A), groundwater level (B), vibration (C), rock and soil elastic modulus (D), temperature (E) and slope top load (F), etc.). Under the influence of basic geological conditions, it is manifested as a time-varying primary term displacement, reflecting the main law of deformation, which is a non-stationary sequence; under the influence of various factors, it is manifested as a randomly changing disturbance term displacement, which is a stationary noise sequence. The existence of this sequence will reduce the authenticity and accuracy of the prediction. As foundation pit projects develop towards larger, deeper and more complex trends, the difficulty of support technology is increasing, and the impact of foundation pit deformation on the surrounding environment is becoming more prominent. In the prediction process, existing models often ignore the interference of external factors on the prediction results, resulting in prediction results that are difficult to meet actual engineering needs. Summary of the invention

[0004] The present invention provides a cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit, so as to overcome the technical problem that the prior art easily ignores the interference of external influencing factors on the prediction results, resulting in inaccurate prediction results of the deformation of the side wall of the foundation pit.

[0005] In order to achieve the above object, the technical solution of the present invention is: A cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit, the specific steps include: S1: Obtain influencing factor data of the inclination degree of the foundation pit side wall and inclination displacement data of the foundation pit side wall; and obtain original displacement data and historical influence data based on the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall, wherein the historical influence data includes disturbance displacement data and influencing factor data of the inclination degree of the foundation pit side wall; S2: establishing a native tilt displacement prediction model, and training it based on the native displacement data to obtain a trained native tilt displacement prediction model; S3: Establish a disturbance tilt displacement prediction model that takes into account the cross-time domain influence characteristics of external factors, and train it based on historical influence data to obtain a trained disturbance tilt displacement prediction model; S4: Based on the trained native tilt displacement prediction model and the trained disturbance tilt displacement prediction model, a total displacement prediction model is obtained, and the foundation pit side wall displacement prediction is realized based on the total displacement prediction model.

[0006] Furthermore, in S3, a disturbance tilt displacement prediction model is established that takes into account the cross-time domain influence characteristics of external factors. The disturbance tilt displacement prediction model includes formulas (1) to (7), and the process of training the model based on historical influence data is as follows: S31: dividing the historical impact data equally according to the time series of foundation pit deformation; and dividing the equally divided historical impact data into a number of calculation intervals according to the detection time; S32: The historical impact data P of the Hth calculation interval is H Substitute the screening series formula to obtain the screened data Q H , expressed as: Q H =X(M Q ×U H-1 +N Q ×P H +G Q ) (1) Among them, X is the sigmoid activation function, M Q and N Q are the coefficients of the linear relationship of the screening series, G Q To screen the error of linear relationship of the series; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; S33: The historical impact data P of the Hth calculation interval is H Substitute the entry sequence formula into it and it becomes: Y H =X(M Y ×U H-1 +N Y ×P H +G Y ) (2) V H =tanh(M V ×U H-1 +N V ×P H +G V ) (3) Where Y H and V HThe entry sequence can store the data used by the model at the best level, X is the sigmoid activation function, M Y 、N Y 、M V and N V is the coefficient of the linear relationship of the input sequence, G Y and G V is the error of the linear relationship of the input sequence; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; S34: Output result V of the H-1th calculation interval based on formulas (1), (2) and (3) H-1 Optimize, the optimization formula is: V H =V H-1 ×Q H +Y H ×V H (4) S35: The data optimized by formula (4) and the historical impact data P of the Hth calculation interval are H Substituting into the export function formula, it is expressed as: I H =X(M I ×U H-1 +N I ×P H +G I ) (5) U H =I H ×tanh(V H ) (6) In the formula, I H is the current data result output by the export sequence, X is the sigmoid activation function, M I and N I is the coefficient of the linear relationship of the export series, G I is the error of the linear relationship of the export series, U H is the abstract expression matrix of the inclination degree of the foundation pit side wall in the Hth calculation interval; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; S36: U H Substituting the result output formula, the predicted result of the disturbance displacement of the pit sidewall inclination on the H+1 day is obtained, which is expressed as: Z * （H+1） =softmax(M S ×U H +G S ) (7) Where MS The relationship coefficient of the result output formula, G S Output the error of the linear relationship for the result.

[0007] Furthermore, in S2, the native tilt displacement prediction model established is: O H =X(M O ×W H +G O ) (8) W * （H+1） =softmax(M W ×O H ) (9) In the formula, O H Optimize the representation coefficient for the model, G O is the average error; W H is the original displacement data, W * （H+1） is the original displacement prediction result on day H+1, M O and M W are the model relationship coefficients respectively.

[0008] Further, in S1, the process of obtaining the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall, and obtaining the original displacement data and the historical influencing data based on the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall is as follows: Construct a monitoring data set J=(J1,J2......J H ); Among them, J1, J2...J H It represents the data of the pit sidewall tilt displacement and influencing factors on day H, and is not limited to consecutive days H+1; J1=(P A 1 ,P B 1 ,P C 1 ,P D 1 ,P E 1 ,P F 1 ,S1), J2=(P A 2 ,P B 2 ,P C 2 ,P D 2,P E 2 ,P F 2 ,S2), ..., J H =(P A H ,P B H ,P C H ,P D H ,P E H ,P F H ,S H ); P A H ,P B H ,P C H ,P D H ,P E H ,P F H They represent the contents of the factors affecting the excavation depth, groundwater level, vibration effect, elastic load of rock and soil, temperature and top load on the Hth day, and are not limited to H consecutive days; S H The data of the inclination displacement of the foundation pit side wall on the Hth day; According to whether the inclination displacement data of the foundation pit side wall is affected by the influencing factors of the foundation pit inclination degree, the inclination displacement data of the foundation pit side wall is divided into original displacement data and disturbed displacement data, and the disturbed displacement data and the influencing factor data of the inclination degree of the foundation pit side wall are combined into historical impact data.

[0009] Furthermore, the total displacement prediction model is expressed as: S * (H+1) =W * (H+1) +Z * (H+1) (10) In the formula, S * (H+1) is the total predicted displacement, W * (H+1) Z is the native displacement predicted by the native tilt displacement prediction model; * (H+1) The disturbance displacement predicted by the disturbance tilt displacement prediction model.

[0010] Beneficial effects: The present invention decomposes the historical foundation pit side wall tilt displacement data, uses the original tilt displacement prediction model to calculate the approximate situation of the on-site foundation pit displacement, and then uses the disturbance tilt displacement prediction model to predict the comprehensive impact of external influencing factors on the foundation pit displacement, and finally uses the total displacement prediction model to obtain the accurate prediction value of the foundation pit deformation displacement. The present invention fully considers the interference of external influencing factors on the prediction results, can scientifically and effectively avoid the occurrence of over-protection and collapse accidents that are common in the foundation pit construction process, saves manpower and material resources for on-site construction, and can also ensure on-site construction safety, which is more in line with the actual application needs of the construction site. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0012] Figure 1 The present invention is a flow chart of a method for predicting the inclination trend of the side wall of a deep and large foundation pit across time domain. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0014] This embodiment provides a cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit. Figure 1 As shown, the specific steps include: S1: Obtain influencing factor data of the inclination degree of the foundation pit side wall and inclination displacement data of the foundation pit side wall; and obtain original displacement data and historical influence data based on the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall, wherein the historical influence data includes disturbance displacement data and influencing factor data of the inclination degree of the foundation pit side wall; In a specific embodiment, in S1, the process of obtaining the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall is as follows: Construct a monitoring data set J=(J1,J2......JH ); In this embodiment, the monitoring data set is derived from daily monitoring of the inclination displacement of the foundation pit side wall.

[0015] Among them, J1, J2...J H It represents the data of the pit sidewall tilt displacement and influencing factors on day H, and is not limited to consecutive days H+1; J1=(P A 1 ,P B 1 ,P C 1 ,P D 1 ,P E 1 ,P F 1 ,S1),J2=(P A 2 ,P B 2 ,P C 2 ,P D 2 ,P E 2 ,P F 2 ,S2),...,J H =(P A H ,P B H ,P C H ,P D H ,P E H ,P F H ,S H ); P A H ,P B H ,P C H ,P D H ,P E H ,P F H They represent the contents of the factors affecting the excavation depth, groundwater level, vibration effect, elastic load of rock and soil, temperature and top load on the Hth day, and are not limited to H consecutive days; S H The data of the inclination displacement of the foundation pit side wall on the Hth day; According to whether the inclination displacement data of the foundation pit side wall is affected by the influencing factors of the foundation pit inclination degree, the inclination displacement data of the foundation pit side wall is divided into original displacement data and disturbed displacement data, and the disturbed displacement data and the influencing factor data of the inclination degree of the foundation pit side wall are combined into historical impact data.

[0016] S2: establishing a native tilt displacement prediction model, and training it based on the native displacement data to obtain a trained native tilt displacement prediction model; In a specific embodiment, in S2, the native tilt displacement prediction model established is: O H =X(M O ×W H +G O ) (8) W * （H+1） =softmax(M W ×O H ) (9) In the formula, O H Optimize the representation coefficient for the model, G O is the average error; W H is the original displacement data, W * （H+1） is the original displacement prediction result on day H+1, M O and M W are the model relationship coefficients respectively.

[0017] S3: Establish a disturbance tilt displacement prediction model that takes into account the cross-time domain influence characteristics of external factors, and train it based on historical influence data to obtain a trained disturbance tilt displacement prediction model; In a specific embodiment, in S3, a disturbance tilt displacement prediction model is established that takes into account the cross-time domain influence characteristics of external factors. The disturbance tilt displacement prediction model includes formulas (1) to (7), and the process of training the model based on historical influence data is as follows: S31: dividing the historical impact data equally according to the time series of foundation pit deformation; and dividing the equally divided historical impact data into a number of calculation intervals according to the detection time; S32: The historical impact data P of the Hth calculation interval is H Substitute the screening series formula to obtain the screened data Q H , expressed as: Q H =X(M Q ×U H-1 +N Q ×P H +G Q ) (1) Among them, X is the sigmoid activation function, M Q and N Q are the coefficients of the linear relationship of the screening series, G Q To screen the error of linear relationship of the series; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; Specifically, in this embodiment, the calculation intervals are divided so that the model can be better trained, and the calculation intervals are used to gradually update and optimize various parameters and gradually transfer the optimized parameters to subsequent calculation intervals, so that various parameters reach the best; and in the process of optimizing the parameters in each calculation interval, it is necessary to first define the prediction of the next data by the previous data, which requires dividing the input information and inputting it into different calculation intervals respectively, so the disturbance displacement input into each calculation interval is different.

[0018] S33: The historical impact data P of the Hth calculation interval is H Substitute the entry sequence formula into it and it becomes: Y H =X(M Y ×U H-1 +N Y ×P H +G Y ) (2) V H =tanh(M V ×U H-1 +N V ×P H +G V ) (3) Where Y H and V H The entry sequence can store the data used by the model at the best level, X is the sigmoid activation function, M Y 、N Y 、M V and N V is the coefficient of the linear relationship of the input sequence, G Y and G V is the error of the linear relationship of the input sequence; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; S34: Output result V of the H-1th calculation interval based on formulas (1), (2) and (3) H-1 Optimize, the optimization formula is: V H =V H-1 ×Q H +Y H ×VH (4) S35: The data optimized by formula (4) and the historical impact data P of the Hth calculation interval are H Substituting into the export function formula, it is expressed as: I H =X(M I ×U H-1 +N I ×P H +G I ) (5) U H =I H ×tanh(V H ) (6) In the formula, I H is the current data result output by the export sequence, X is the sigmoid activation function, M I and N I is the coefficient of the linear relationship of the export series, G I is the error of the linear relationship of the export series, U H is the abstract expression matrix of the inclination degree of the foundation pit side wall in the Hth calculation interval; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; Specifically, in this embodiment, the abstract expression matrix of the initial inclination of the foundation pit side wall is an artificially set value, which is subsequently further optimized by formula (6).

[0019] S36: U H Substituting the result output formula, the predicted result of the disturbance displacement of the pit sidewall inclination on the H+1 day is obtained, which is expressed as: Z * （H+1） =softmax(M S ×U H +G S ) (7) Where M S The relationship coefficient of the result output formula, G S Output the error of the linear relationship for the result.

[0020] Specifically, in this embodiment, the disturbance inclination displacement prediction model filters and stores input information through a screening series and an entry series, thereby being able to effectively eliminate the hysteresis effect of the factors affecting the inclination degree of the foundation pit sidewall on the inclination displacement of the foundation pit sidewall.

[0021] S4: Based on the trained native tilt displacement prediction model and the trained disturbance tilt displacement prediction model, a total displacement prediction model is obtained, and the foundation pit side wall displacement prediction is realized based on the total displacement prediction model.

[0022] In a specific embodiment, the total displacement prediction model is expressed as: S * (H+1) =W * (H+1) +Z * (H+1) (10) In the formula, S * (H+1) is the total predicted displacement, W * (H+1) Z is the native displacement predicted by the native tilt displacement prediction model; * (H+1) The disturbance displacement predicted by the disturbance tilt displacement prediction model.

[0023] In this embodiment, after the total displacement prediction model in the present invention is trained based on the foundation pit side wall inclination related data in Table 1, the deformation of the foundation pit is predicted based on the trained total displacement prediction model. In Table 1, the side wall inclination displacement values ​​in the last column are derived from the monitoring of a series of on-site equipment and facilities.

[0024] Table 1:

[0025] Specifically, we construct a monitoring data set J=(J1,J 2, ...,J 100 ), which is a dataset derived from the monitoring of the sidewall inclination displacement of the foundation pit in the experimental area of ​​the Renqiu cable-stayed bridge.

[0026] in: J 94 =(P A 94 ,P B 94 ,P C 94 ,P D 94 ,P E 94 ,P F 94 ,S 94 ), J 95 =(P A 95 ,P B 95 ,P C 95 ,P D 95 ,P E 95,P F 95 ,S 95 ), ..., J 100 =(P A 100 ,P B 100 ,P C 100 ,P D 100 ,P E 100 ,P F 100 ,S 100 ), Among them: J 94 ,J 95 ......J 100 Represents the monitoring data of the last 7 days and is not limited to 7 consecutive days; P A 100 ,P B 100 ,P C 100 ,P D 100 ,P E 100 ,P F 100 They represent the excavation depth (A), groundwater level (B), vibration impact (C), rock and soil elastic load (D), temperature (E) and slope top load (F) on the 100th day. The units of each factor are m, m, Hz, MN / m 2 , o C. KN / m.

[0027] The monitoring displacement is decomposed by using existing decomposition methods such as variational mode decomposition theory and wavelet optimization decomposition theory, and the original displacement data and disturbance displacement data are obtained as shown in Table 2 below. The disturbance displacement prediction model training data are constructed respectively, which is expressed as: L1=(P A 1 ,P B 1 ,P C 1 ,P D 1 ,P E 1 ,P F 1 ,Z1),P2=(P A 2 ,P B 2 ,PC 2 ,P D 2 ,P E 2 ,P F 2 ,Z2),..., L 100 =(P A 100 ,P B 100 ,P C 100 ,P D 100 ,P E 100 ,P F 100 ,Z 100 ) and the native displacement prediction model training data, expressed as: W1,W2,...,W 100Z。

[0028] Table 2:

[0029] Specifically, this embodiment takes the 100th set of data as an example, and brings the original displacement of the foundation pit into the original tilt displacement prediction model, which is expressed as: , , The foundation pit disturbance displacement and influencing factors are introduced into the disturbance tilt displacement prediction model, which is expressed as: , , , , , U H =I H ×tanh(V H )=1.19×tanh(30.82)=5.79, Z * （H+1） =softmax(M S ×U H +G S )=softmax(0.21×5.79+0.04)=0.99, S * (H+1) =W *(H+1) +Z * (H+1) =0.99+16.49=17.4.

[0030] Specifically, Table 3 summarizes the prediction errors of the test sample group. It can be seen from Table 3 that the maximum absolute error of the prediction results using the method proposed in the present invention is 0.43, the maximum relative error is 2.3%, and the accuracy is relatively high, which can meet the prediction and evaluation requirements of the inclination trend of the foundation pit side wall. Therefore, it is explained that the total displacement prediction model proposed in this embodiment can directly, quickly and accurately calculate the inclination displacement of the foundation pit side wall based on the currently measured excavation depth (A), groundwater level (B), vibration impact (C), rock and soil elastic modulus (D), temperature (E) and slope top load (F), without the need for a large number of complicated evaluations and calculations, which greatly improves the efficiency of the prediction and evaluation calculation of the inclination trend of the foundation pit side wall.

[0031] Table 3: In summary, the present invention can predict the deformation and displacement of the foundation pit based entirely on the two models, thereby realizing the determination of the stability of the foundation pit. It is an intelligent digital prediction method and improves the accuracy of the calculation results. Specifically, it also includes the following beneficial effects: (1) Data collation and statistics of historical monitoring data sets: This embodiment inputs the data of field monitoring into the overall model, which can automatically collate and summarize the monitoring data. Traditional monitoring data relies on abstract EXCEL tables or even paper files, which are prone to data loss, management confusion and other problems. The present invention can summarize the data through the model, avoiding the problem of data loss and confusion; (2) Accurate prediction of foundation pit deformation and displacement: An important indicator for determining foundation pit stability is the deformation and displacement of the foundation pit sidewall at the current moment. The present invention constructs a native tilt displacement prediction model and a disturbance tilt displacement prediction model. The construction personnel only need to input the historical monitoring data set into the model to quickly obtain the predicted value of the foundation pit deformation and displacement at the next moment, thus avoiding dependence on the technical experience of the construction personnel and having high accuracy. (3) Treatment of time lag of external factors: In the prediction process of disturbance tilt displacement, the disturbance tilt displacement prediction model uses three-term series to save the prediction results of the input information in each previous interval, which effectively solves the time lag problem of external factor response displacement. Compared with the existing methods, this method is more stable, gives full play to the advantages of model calculation, and the prediction results are more accurate.

[0032] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting the inclination trend of the sidewall of a deep and large foundation pit across time domain, characterized in that: The specific steps include: S1: Obtain influencing factor data of the inclination degree of the foundation pit side wall and inclination displacement data of the foundation pit side wall; and obtain original displacement data and historical influence data based on the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall, wherein the historical influence data includes disturbance displacement data and influencing factor data of the inclination degree of the foundation pit side wall; S2: establishing a native tilt displacement prediction model, and training it based on the native displacement data to obtain a trained native tilt displacement prediction model; S3: Establish a disturbance tilt displacement prediction model that takes into account the cross-time domain influence characteristics of external factors, and train it based on historical influence data to obtain a trained disturbance tilt displacement prediction model; S4: Based on the trained native tilt displacement prediction model and the trained disturbance tilt displacement prediction model, a total displacement prediction model is obtained, and the foundation pit side wall displacement prediction is realized based on the total displacement prediction model.

2. The cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit according to claim 1 is characterized in that: In S3, a disturbance tilt displacement prediction model is established that takes into account the cross-time domain influence characteristics of external factors. The disturbance tilt displacement prediction model includes formulas (1) to (7), and the process of training it based on historical influence data is as follows: S31: dividing the historical impact data equally according to the time series of foundation pit deformation; and dividing the equally divided historical impact data into a number of calculation intervals according to the detection time; S32: The historical impact data P of the Hth calculation interval is H Substitute the screening series formula to obtain the screened data Q H , expressed as: Q H =X(M Q ×U H-1 +N Q ×P H +G Q ) (1) Among them, X is the sigmoid activation function, M Q and N Q are the coefficients of the linear relationship of the screening series, G Q To screen the error of linear relationship of the series; U H-1 is the abstract expression matrix of the inclination degree of the foundation pit side wall in the H-1th calculation interval; S33: The historical impact data P of the Hth calculation interval is H Substitute the entry sequence formula into it and it becomes: Y H =X(M Y ×U H-1 +N Y ×P H +G Y ) (2) V H =tanh(M V ×U H-1 +N V ×P H +G V ) (3) Where Y H and V H The entry sequence can store the data used by the model at the best level, X is the sigmoid activation function, M Y 、N Y 、M V and N V is the coefficient of the linear relationship of the input sequence, G Y and G V is the error of the linear relationship of the input series; S34: Output result V of the H-1th calculation interval based on formulas (1), (2) and (3) H-1 Optimize, the optimization formula is: V H =V H-1 ×Q H +Y H ×V H (4) S35: The data optimized by formula (4) and the historical impact data P of the Hth calculation interval are H Substituting into the export function formula, it is expressed as: I H =X(M I ×U H-1 +N I ×P H +G I ) (5) U H =I H ×tanh(V H ) (6) In the formula, I H is the current data result output by the export sequence, X is the sigmoid activation function, M I and N I is the coefficient of the linear relationship of the export series, G I is the error of the linear relationship of the export series, U H is the abstract expression matrix of the inclination degree of the foundation pit side wall in the Hth calculation interval; S36: U H Substitute the result output formula to obtain the disturbance displacement prediction result Z of the inclination degree of the foundation pit side wall on the H+1 day * （H+1） , expressed as: Z * （H+1） =softmax(M S ×U H +G S ) (7) Where M S The relationship coefficient of the result output formula, G S Output the error of the linear relationship for the result.

3. The cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit according to claim 2 is characterized in that: In S2, the original tilt displacement prediction model established is: O H =X(M O ×W H +G O ) (8) W * （H+1） =softmax(M W ×O H ) (9) In the formula, O H Optimize the representation coefficient for the model, G O is the average error; W H is the original displacement data, W * （H+1） is the original displacement prediction result on day H+1, M O and M W are the model relationship coefficients respectively.

4. The cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit according to claim 3 is characterized in that: In S1, the process of obtaining the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall, and obtaining the original displacement data and the historical influence data based on the influencing factor data of the inclination degree of the foundation pit side wall and the inclination displacement data of the foundation pit side wall is as follows: Construct a monitoring data set J=(J1,J2......J H ); Among them, J1, J2...J H It represents the data of the pit sidewall tilt displacement and influencing factors on day H, and is not limited to consecutive days H+1; J1=(P A 1 ,P B 1 ,P C 1 ,P D 1 ,P E 1 ,P F 1 ,S1), J2=(P A 2 ,P B 2 ,P C 2 ,P D 2 ,P E 2 ,P F 2 ,S2), ..., J H =(P A H ,P B H ,P C H ,P D H ,P E H ,P F H ,S H ); P A H ,P B H ,P C H ,P D H ,P E H ,P F H They represent the contents of the factors affecting the excavation depth, groundwater level, vibration effect, elastic load of rock and soil, temperature and top load on the Hth day, and are not limited to H consecutive days; S H The data of the inclination displacement of the foundation pit side wall on the Hth day; According to whether the inclination displacement data of the foundation pit side wall is affected by the influencing factors of the foundation pit inclination degree, the inclination displacement data of the foundation pit side wall is divided into original displacement data and disturbed displacement data, and the disturbed displacement data and the influencing factor data of the inclination degree of the foundation pit side wall are combined into historical impact data.

5. The cross-time domain prediction method for the inclination trend of the side wall of a deep and large foundation pit according to claim 4 is characterized in that: The total displacement prediction model is expressed as: S * (H+1) =W * (H+1) +Z * (H+1) (10) In the formula, S * (H+1) is the total predicted displacement, W * (H+1) Z is the native displacement predicted by the native tilt displacement prediction model; * (H+1) The disturbance displacement predicted by the disturbance tilt displacement prediction model.

Images